OpenAI's Sol model autoformalized a ChatGPT-disproved Erdős Unit Distance conjecture in Lean, generating 1.2 million lines of code in three weeks. Boris Alexeev of OpenAI steered the model to a comple…
UC Berkeley assistant teaching professor Phillip Kerger used OpenAI's GPT-5.6 Sol Pro to derive a new lower bound in convex optimization, then formalized the result in the Lean theorem prover. The 36-…
A team built a prototype that verifies code migration from F# to Python using a proof in the Lean theorem prover, rather than relying on tests. The system translates both the original and rewritten co…
A user on Hacker News asks for functional search engines for mathematical formulas, noting that existing tools like approach0.xyz, searchonmath.com, and mathweb.org are offline or produce irrelevant r…
A new position paper on using formal methods for AI security focuses on model weight confidentiality and integrity through infrastructure hardening, with a minimal and uncontroversial approach. The UK…
ZkSecurity released zk.golf, a platform for competitive optimization of zero-knowledge circuits using formal verification and AI. The platform combines Clean, a tool for machine-checkable proofs of ci…
A new study on natural-language-to-Lean formalization reveals that compile-rate evaluation overestimates faithfulness, with a tool-augmented agent achieving 89.5% compilation but only 60.5% consensus …
Researchers at the University of Washington released TheoremGraph, a unified dependency graph spanning 18 million+ mathematical statements from arXiv papers and the Lean formal proof assistant. The gr…
Abram Demski reports using Claude Opus 4.8 and GPT 5.5 to conduct 'vibe research' on logical induction as a model for AI trustworthiness and recursive self-improvement, collaborating with Anson Berns …
Reinforcement Learning with Verifiable Rewards (RLVR) allows language models to bootstrap beyond human-level capabilities on exactly graded problems like coding and formal proofs, but alignment-flavor…
Cajal (YC W26) released Talos, an open-source WebAssembly interpreter and formal verification framework built in Lean, enabling mathematical proofs of correctness for Wasm modules. The tool targets AI…
John D. Cook used Anthropic's Claude (Sonnet 4.6 Medium) to find a typo in a blog post about quaternion rotations by asking it to write Lean code verifying the post's theorems. After four iterations, …
Researchers introduced Pythagoras-Prover, a family of compute-efficient Lean theorem provers that achieve state-of-the-art performance with significantly fewer parameters. The 4B-parameter model surpa…
Anthropic's Claude AI generated a Lean formal proof for a Fourier coefficient calculation involving Bessel functions, requiring eight iterations to fix errors and produce a working proof with four 'so…
Researchers have developed LeanMarathon, a multi-agent system that reliably autoformalizes complex mathematical proofs into the Lean theorem prover by breaking long-horizon tasks into parallel, recove…
DeepMind's AI agent autonomously resolved 9 of 353 open Erdős problems and proved 44 of 492 open conjectures from the OEIS at an inference cost of a few hundred dollars per problem, using a generate-a…
A developer has released what they claim is the first formally verified implementation of a polygon intersection algorithm, using the Lean proof checker to guarantee correctness. The project, which su…
Researchers have demonstrated that AI agents using large language models to generate formal proofs in Lean can autonomously solve open mathematics problems, resolving 9 of 353 unsolved Erdős problems …
OpenAI's GPT-next model disproved the 80-year-old Erdős planar unit distance problem in under 32 hours at a cost of less than $1,000, marking the first instance of a general-purpose AI solving a well-…
Weekly roundup of the author's reading, covering three main topics: the challenges of using formal languages like Lean for AI-driven mathematical proofs, the history of the GLP-1 drug class and the bi…